Welcome to Runtime! Today: why companies building generative AI applications might not need to wait for perfection, The Linux Foundation gets bigger, and the latest enterprise moves.

(Was this email forwarded to you? Sign up here to get Runtime each week.)

The error era

LAS VEGAS - Companies that have struggled to deploy production generative AI apps over the last year or so report getting somewhere around 80% to 90% through the process before tripping up while applying the finishing touches, mostly because of concerns about accuracy. But is perfection the enemy of good?

That was the sentiment from a panel of data experts moderated by yours truly on Wednesday at the HumanX conference. After a bit of back and forth over the particulars, the group — Mike Gozzo of Ada, Edo Liberty of Pinecone, Barr Moses of Monte Carlo, and Shannon Scott of Airwallex — landed on the sentiment that for most applications, something that works 80% of the time is probably fine.

• "You're never going to be perfect. You're never going to be 100% sure that it's never going to make a mistake," Liberty said, noting that human beings also have a long track record of making mistakes.
• A lot of companies experimenting with generative AI applications started out by building internal apps, reasoning that if something went awry it at least wasn't damaging their brand with customers, but that approach has held people back, Moses said.
• "I'm actually finding that people are not excited to do internal use cases, because there's a fear that you are being replaced. There's a lot more momentum around externally facing products, whether that's revenue generation products or anything that really accelerates sort of personalization," she said.
• According to Liberty, users of generative AI applications are increasingly willing to tolerate a small amount of error if the overall results are still pretty good, "They don't penalize you for being a little bit wonky sometimes because they know what's happening."

That's not true for everyone, of course, and most generative AI app builders want to improve the quality of their apps. But the best way to get better is the same way software developers have always improved their apps; launching, getting feedback, and iterating, which is what Sierra co-founder and CEO Bret Taylor told prospective customers earlier last month.

• "Run your AI project for enough time to see where it breaks, analyze every failure and account for it until it passes. I think we have to just get used to that iterative style of working," Gozzo said.
• Gozzo also pointed out, however, that everyone involved — especially the buyer — needs to have an understanding of what to expect from an 80% app, and that "the cost of those failures isn't exorbitantly high."
• Scott agreed: "If I can keep the scope of the application relatively narrow, it makes it much easier for me to be able to put controls in to assess, is it working?"

Enterprise software companies are making an enormous bet this year that agentic AI, based around newer chain-of-thought or "reasoning" AI models, will help companies deliver more sophisticated apps with better accuracy than early experiments. But when customers kick the tires on an AI agent, they quickly realize that they're not going to make progress until they organize their data.

• "The only way this works is when people start to really think about how to make their internal proprietary data available to those models in real time and make that stack work for them," Liberty said.
• Moses agreed: "It is early days, and it's obviously a lot harder with unstructured data. But people are investing in that diligence in order to increase that confidence."
• Right now buyers have a dizzying array of options when it comes to implementing AI agents, but to a person the panel strongly believed that someone will figure out a way to get past the concerns of naturally skeptical buyers and start delivering results.
• "These services are going to become a basic expectation," Scott said. "The user interface is evolving in a very visceral way for what used to be very structured applications."

I heard you liked foundations…

It's fair to say that OpenStack never lived up to the lofty expectations laid out 15 years ago by its founders, who hoped to build an open-source infrastructure blueprint that would allow companies to get cloud-like performance inside their own data centers. But right as the private cloud community goes through a period of upheaval thanks to Broadcom's 2023 purchase of VMware, OpenStack has found a new home.

The Open Infrastructure Foundation announced this week that it is merging with the Linux Foundation. As TechCrunch noted, "three of the world’s largest and most active open source projects (Linux, Kubernetes, and OpenStack), now fall under the Linux Foundation umbrella."

That decision only further cements the influence that the Linux Foundation — a very corporate-but-we're-not-a-corporation non-profit organization — has over enterprise tech. A lot of companies that want or need to manage their own servers are considering new options following Broadcom's price hikes for VMware's software, and OpenStack now has support from some deep pockets.

Enterprise moves

Lip-Bu Tan is the new CEO of Intel, less than a year after he resigned from Intel's board and at a time when the survival of one of the most iconic tech companies in American history is in doubt.

Maha Virudhagiri is the new CTO of Coalition, joining the cybersecurity insurance company in a newly created role after seven years at Tesla.

David Sudbey is the new chief customer officer at Dialpad, following similar sales leadership roles at Cognito and Genesys.

Scott Fuselier is the new chief revenue officer at Expel, joining the MDR security company following similar roles at Immuta and Menlo Security.

The Runtime roundup

Google Cloud reorganized its sales organization last month in order to "respond faster to market demands," according to an internal email obtained by Business Insider.

Uipath's stock fell more than 15% Thursday after it missed revenue estimates and forecast slower growth for the year ahead, as AI continues to upend the RPA market.

Thanks for reading — see you Saturday!